Application of Artificial Neural Networks for Seismic Analysis and Design of Buried Pipelines in Heterogeneous Soils

Every year, the phenomenon of earthquake causes a lot of human, financial and environmental losses. Transmission pipelines are one of the vital arteries that are very important, however, in the event of an earthquake can cause devastating damages. Safeguarding urban and interurban facilities, including electricity, water supply, oil and gas transmission lines, against these loads requires careful studies and engineering designs. Given that traditional methods for seismic design of pipelines such as FEM modeling and experimental methods are so expensive, a new combined method for predicting the strain of pipes based on the Artificial Neural Network (ANN) is proposed. For this purpose, the parameters of the pipeline including pipe and soil type, length, discharge, path slope, depth, etc. and earthquake-induced characteristics including earthquake acceleration, earthquake occurrence time, Peak ground acceleration (PGA), etc. were included in the model. Earthquake input parameters were considered as input parameters and pipe strains and stresses were considered as output parameter. ANSYS finite element software has also been used to simulate the pipeline and produce training data. The results of finite element software were used as input and output parameters for training and validating artificial neural network. 753 models created using ANSYS and its input/output data divided into three parts to creat ANN model. 70% of the total data were used for training, 15% for validating and 15% for testing the ANN model. Results show that the proposed Method provides a very good agreement with the computational results of the ANSYS with accuracy of 96 percent.